The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The script scripts/generate_prd.sh is vulnerable to arbitrary command execution through the unsafe use of eval.
Evidence: The prompt_input function in scripts/generate_prd.sh uses eval "$var_name='$input'" to assign values to variables. An attacker providing input containing shell metacharacters (e.g., '; touch /tmp/pwned; #) can execute arbitrary commands with the privileges of the user running the script.
[COMMAND_EXECUTION] (MEDIUM): The script scripts/generate_prd.sh allows for potential arbitrary file writes due to lack of path sanitization.
Evidence: The OUTPUT_FILE variable is constructed directly from the user-provided PRODUCT_NAME using ${PRODUCT_NAME// /_}_prd.md. An attacker can use path traversal sequences (e.g., ../../../tmp/evil) in the product name to write files outside of the intended directory.
[INDIRECT_PROMPT_INJECTION] (LOW): The validation script scripts/validate_prd.sh provides a surface for indirect prompt injection by processing potentially untrusted PRD files.
Ingestion points: scripts/validate_prd.sh reads and parses the file provided in the first argument ($PRD_FILE).
Boundary markers: None. The script performs raw grep operations on the file content.
Capability inventory: The script uses grep and wc to analyze the file and prints the results to stdout, which may be consumed by an AI agent.
Sanitization: No sanitization or validation of the input file's content is performed before analysis.
[DATA_EXPOSURE] (LOW): The script scripts/generate_prd.sh automatically includes system information in the generated document.
Evidence: The script executes whoami and date to populate the 'Author' and 'Date' fields in the PRD template, exposing the current username and system time.

prd-generator